ultralytics-segment-json
All format examples · Machine-readable schema · Convert outputs to COCO
One image, one instance; normalize=False. Corners and contours use original-image pixels. Image identity and dimensions come from separate context.
Example
Illustrative values, not predictions from a model. Unless stated otherwise, the image is 200 pixels wide and 100 pixels high.
[
{
"box": {
"x1": 10,
"x2": 50,
"y1": 20,
"y2": 60
},
"class": 0,
"confidence": 0.8,
"name": "insect",
"segments": {
"x": [10, 50, 50, 10],
"y": [20, 20, 60, 60]
}
}
]Convert to a standard format
This format is documented, but a conversion adapter is not yet registered.
General conversion example and required context
How to interpret it
Scope: One image per call; top-level array of instances. No image path, dimensions or class-name lookup table in the JSON.
Coordinates: Default: original-image pixel corner coordinates. normalize=True divides x by image width and y by height, including segmentation coordinates.
Tracking: track_id is omitted for ordinary predictions; present for tracking results.
Empty: No detections produces an empty array.